The degree to which the high species richness of Mediterranean-climate shrublands is maintained by competition-driven differentiation of functional traits [11, 20], or by lottery recruitment and disturbance-mediated coexistence [14, 52], is not fully clear. These two models make different predictions about the associations between species co-occurrence, phylogenetic relatedness, and phenotypic similarity. I have shown that for Banksia in southwestern Australia, these associations vary between fire-killed and fire-resistant species, suggesting that the ecological processes that govern co-occurrence and community structure do not necessarily apply uniformly to all species within the same communities.
The three scenarios entertained in the Introduction were that (1) patterns for resprouters but not reseeders are consistent with niche differentiation; (2) patterns for reseeders but not resprouters are consistent with niche differentiation; (3) there are no differences in patterns between resprouters and reseeders. The results provide greatest support for scenario 2, with reseeders, but not resprouters, showing patterns of phylogenetic and phenotypic repulsion consistent with niche differentiation. Of the traits included in this study, maximum adult height appears to play a role in driving the phylogenetic signal of co-occurrences among reseeders, as it shows a pattern of phenotypic repulsion and is strongly phylogenetically conserved. In contrast, the results provide no evidence that seed size, or the length or degree of overlap in flowering period, are associated with patterns of co-occurrence. The adult height attained by a plant species represents a tradeoff between the benefits of greater height, such as greater light interception and seed production, and the costs, such as greater proportional investment in non-reproductive support tissue [24, 25]. Height is therefore a key dimension of ecological variation in plants  and a potential axis for niche differentiation where competition between species exists [53, 54]. The plots of contrasts in height against node ages show no evidence for rapid divergence among closely-related lineages. Height therefore appears to influence the co-occurrence patterns of reseeding Banksia species by ecological sorting (i.e. competitive exclusion) rather than through microevolutionary divergence.
Why should there be evidence of niche differentiation along the axis of height for reseeders, but not for resprouters? There is likely to be selective pressure on reseeder species to grow rapidly and reach full seed production within the average fire recurrence period, favouring a lower adult height . This is traded off against the costs of faster growth and lower adult height, including greater soil nutrient requirements, lower total seed production and reduced seed dispersal capacity [22, 25, 53]. On the other hand, it is less easy to explain why these selective pressures should be absent in resprouters. It has been observed that some resprouting Banksia species have very low rates of recruitment from seed, and most plants present after a fire are adults that have resprouted, rather than new recruits [23, 55]. Furthermore, adult height does not reflect longevity in resprouters as it does in reseeders. Hence, the costs and benefits of rapid growth and lower height may be less critical for resprouters compared to reseeders. Of course, it is also possible that because of the smaller number of pairs of resprouter species (48 compared to 174 reseeder pairs), there was simply less power to detect patterns of phylogenetic and phenotypic repulsion. Another possibility is that there may be other niche dimensions not included in my study that influence co-occurrence patterns among resprouters.
The patterns of phenotypic repulsion based on height are set against a background of environmental filtering, whereby species with similar edaphic preferences (with respect to soil nitrate concentration and pH) are more likely to co-occur. This is consistent with the high degree of specialization to soil types that seems to characterize old, nutrient-impoverished landscapes in southwest Australia and elsewhere . However, it is not obvious why Banksia co-occurrences should be associated with soil nitrate rather than with phosphorus, which is more likely to be the limiting nutrient in the highly weathered soils of southwestern Australia [56, 57]. One possibility is that the measures of phosphorus concentration used here were for total phosphorus rather than plant-available forms. Unlike height, species’ edaphic preferences are highly labile, so this pattern is essentially independent of relatedness, and thus leaves no signature in the phylogenetic patterns of co-occurrence. The lability of edaphic preferences is consistent with the hypothesis that local-scale specialization to soil types happens rapidly, potentially contributing to rapid speciation rates in Mediterranean-climate shrublands [14, 15, 58].
My approach to analyzing community phylogenetic patterns based on pairwise co-occurrences is similar to that employed for schoenoid sedges in another Mediterranean-climate shrubland, the Cape Floristic Region, which also found evidence for phylogenetic repulsion . This approach differs from a recent study by Merwin et al. that analyzed phylogenetic structure in Banksia communities using whole-assemblage metrics. In contrast to my finding of phylogenetic repulsion among species pairs, Merwin et al. found that many communities were phylogenetically clustered, indicating that closely-related species are more likely to co-occur. This contrast can most likely be explained as an issue of spatial scale. Their plots spanned a far greater geographical area than mine, and they interpret phylogenetic clustering as the signal of speciation and limited dispersal abilities generating many closely-related but narrowly-distributed species within a given region. In contrast, not only were my plots spread over a smaller area, but I explicitly attempted to minimize the signal of speciation history by omitting non-overlapping pairs of species from the analysis. In this way, my analyses of co-occurrence patterns were more likely to have detected local-scale ecological patterns rather than broad biogeographic effects.
Similarly, my finding of phylogenetic repulsion appears at odds with another recent study that showed that fire regeneration mode can drive phylogenetic clustering through shared adaptation of close relatives to fire-prone environments . Again, the difference can be explained by a difference in scale: the study by Verdu & Pausas  examined patterns across two major habitat types, one of which was fire-prone and the other not. At these scales, environmental filtering is likely to predominate over competition; furthermore, fire response in the flora they examined was phylogenetically conserved. This contrasts with Banksia in which regeneration mode is highly labile and its primary influence on community structure appears to be by mediating competitive interactions among species.